Sep 20 – 25, 2026
University of Graz
Europe/Vienna timezone

Machine-Learning Potentials for Nano-Mechanical Simulations of Defective Ceramics: A Case Study of Transition Metal Diborides

Sep 25, 2026, 11:00 AM
15m
HS 12.01 (University of Graz)

HS 12.01

University of Graz

12 - Heizhaus, ground floor
3) Contributed talk OGD: Surfaces, Interfaces and Thin Films Parallel

Speaker

Chunhui Du (TU Wien)

Description

Transition metal diborides (TMB$_2$s) ceramics are highly attractive for protective coating applications due to their high hardness with excellent thermal and chemical stability. Synthesis conditions cause these materials commonly grown as largely off-stoichiometric, consequently, including vacancies or other simple crystallographic defects that alter the intrinsic response to mechanical strains. Therefore, it is now critical to establish an atomic-level understanding of how such defects control the mechanical properties of TMB$_2$s ceramics.

In this work, molecular dynamics (MD) simulations equipped with here-trained machine-learning interatomic potentials (MLIP) are carried out, to reveal effects of point and planar defects during typical mechanical loading of several paradigm diborides (TMB$_2$, M $=$ Ti, Ta and W). MLIP training routine consists of active learning on configurations from 0K DFT calculations and finite temperature ab initio molecular dynamics, including equilibrium structures of different phases ($\alpha$, P6/mmm and $\omega$, P6$_3$/mmc), uniaxial/shear loading states, as well as various defective and/or extremely strained environments. The MLIP's robustness to highly strained environments, particularly near indenter tips, is achieved via on-the-fly training on extrapolative atomistic clusters from simple nanoindentation runs.
Following MLIP's validation, we simulate room-temperature tensile tests and nanoindentation of TMB$_{2}$ and TMB$_{2\pm x}$ structures, where TM/B sub-stoichiommetry is realized by disordered TM/B vacancies and planar defects previously observed by electron microscopy.

The results demonstrate that our trained MLIP remain robust throughout all simulations. It also accurately captured phase transformations under various loading conditions. A particularly surprising and non-intuitive prediction is that some non-stoichiometric TMB$_{2\pm x}$ structures with specific defect types can exhibit even higher hardness and Young's modulus comparable to the stoichiometric TMB$_{2}$, challenging traditional assumptions about weakening effects of sub-stoichiometry. Moreover, this study elucidates the correlation between formation energies and mechanical responses of these non-stoichiometric structures. Overall, these findings offer valuable insights for advanced defect engineering strategies in the future development of transition metal diboride (TMB$_x$) thin films.

Author

Chunhui Du (TU Wien)

Co-authors

Prof. Davide Sangiovanni (Linköping University) Prof. Nikola Koutná (TU Wien) Paul Mayrhofer (TU Wien) Dr Shuyao Lin (TU Wien)

Presentation materials

There are no materials yet.